VideoPoet by Google

VideoPoet by Google is a large language model for video that generates and transforms video content directly from text, images, or existing footage. Built as a unified multimodal model, it supports text-to-video, image-to-video, video stylization, and video editing within a single framework. Users can create short video clips from written prompts, extend or modify existing videos, or animate static images by describing desired motion, style, or atmosphere.

The model operates autoregressively on visual and audio tokens, enabling synchronized generation of both video frames and sound. This allows VideoPoet to produce coherent, temporally consistent motion and basic audio that align with the input description. It can perform tasks such as adding motion to still images, changing the visual style of a clip, performing subject-driven edits, or generating looping or seamlessly transitioning video segments.